From Digital Twin to Competitive Advantage: Designing a Digital Thread-Enabled Predictive Quality Data Ecosystem for Aerospace Manufacturing

The competitive advantage in aerospace manufacturing is more and more based on ensuring high conformance at high throughput and audit-ready traceability through complex supply networks. Advanced technologies like laser powder bed fusion (PBF-LB/M) generate rich multi-modal data (manufacturing intent, execution logs, in-situ monitoring, and ex-situ verification), but value is often compromised by fragmented repositories, weak semantic/coordinate alignment, and limited governance of analytics. In this study, we first present a Digital Thread-enabled Predictive Quality Data Ecosystem (PQDE) that leverages digital twin capabilities (state estimation and quality-risk forecasting) with measurable operational outcomes. PQDE combines (i) digital-thread continuity from manufacturing intent to inspection evidence through identifier binding and multi-modal registration, (ii) reproducible DataOps and MLOps controls (schema, lineage, acceptance gates, monitoring and controlled changes), and (iii) decision integration through risk-guided inspection triage with evidence packaging in alignment with aerospace quality expectations. This approach is validated with the public NIST Additive Manufacturing Metrology Testbed (AMMT) pilot “Overhang Part X4,” which contains a fully registered in-situ/ex-situ dataset, where all of this data is matched with machine coordinates and XCT-derived verification. By leveraging a governance-friendly baseline model and a rare-event label created based on the 1st percentile threshold of the training split, we obtain a strong level of discrimination among validation and cross-part test slices (ROC-AUC 0.953-0.982; PR-AUC up to 0.448). More specifically, Top–10% triage captures 98.20% of defect-labeled points for validation and 96.88% for late-layer test slice while only reducing the inspection scope by 90%; Top–10% triage captures 83.07% on that more challenging early-layer test slice. Lift values of ~8.31x – 23.82x reflect the significant defect enrichment in inspected subsets. Finally, it demonstrates KPI proxy templates (inspection reduction, defect capture, enrichment, false-alarm load, decision-latency proxy, and evidence completeness) to transfer the predictive outputs into competitive advantage without needing proprietary cost-of-quality data.